Research On Image Super-resolution Reconstruction Algorithm Based On Local Edge Feature Constraint

Image super-resolution reconstruction technology has been widely used in video surveillance,satellite remote sensing,military reconnaissance,medical image processing and other fields.Generally,the problem of image degradation exists widely in various scenes,and the factors leading to image damaging are very complex and have great uncertainty,which makes the research on image super-resolution reconstruction technology become an urgent need.The difficulty of image super-resolution reconstruction problem lies in that the high-frequency information of the image is hard to recover due to the lack of enough prior information.Moreover,the reconstruction method usually has the problems of complex calculation and low real-time performance.In this thesis,the sparse representation method is used to study the problem of super-resolution reconstruction of single-frame low-resolution images without noise,and an image super-resolution(LEF-SCSR)algorithm based on local edge feature constraint is proposed.The work of this thesis includes two parts: the first part is about the research on local edge constraint,the other is about some research on the global optimization method.Firstly,the LEF-SCSR algorithm is preliminarily studied and the sparse coding module is optimized by adopting the multi-dictionary joint training strategy and introducing the local edge feature constraint.Sparse coding module is the key part of image SR algorithm based on sparse representation.It directly determines the quality of reconstructed HR image,and unsatisfactory sparse coding results will lead to serious distortion of reconstructed HR image.The LEF-SCSR algorithm,based on the sparsity of image features such as edge texture,can better suppress the generation of edge jagged and artifact blur,and effectively restore the high-frequency details of the image,and finally achieve improvement in visual effect and objective measurement.In the second part,a global optimization method for LEF-SCSR algorithm is proposed by combining the asymptotic up sampling networks and iterative up and down sampling networks.The proposed network can fully mine and utilize the information contained in the low resolution image,and recover the high-frequency details of the image as much as possible.The enhanced reconstruction constraint condition is realized by applying the reconstructed edge feature image to the reconstruction constraint module,which can keep the high frequency information of the reconstructed image from distortion in the iterative process.The experimental results suggest that the edge sawtooth effect in the restored image by LEF-SCSR algorithm is significantly reduced compared with the method before optimization.And the reconstructed image has sharper edge and texture feature,which finally achieves further improvements in visual effects and objective metrics.